Job offer

The position is hosted by the ETIS Lab (UMR 8051), a joint research unit of CY Cergy Paris Université, ENSEA and CNRS, located in Pontoise in the Cergy-Pontoise area near Paris. The researcher will join the team working on data science, information systems and knowledge management, under the supervision of Professor Dimitris Kotzinos, and will collaborate closely with the European partners of the ECHOES consortium, with opportunities for travel, project meetings and research visits across Europe.

• PhD in Computer Science or a closely related field, obtained before the starting date.
• Strong background in at least one of: machine learning on graphs (GNNs, KG embeddings), federated/distributed learning, knowledge graphs and Semantic Web technologies (RDF, SPARQL, OWL, CIDOC-CRM is a plus), or distributed data management.
• Solid programming skills (Python; PyTorch/PyTorch Geometric, DGL or similar; experience with triple stores or graph databases is appreciated).
• A track record of publications in recognised international venues.
• Ability to work in a multidisciplinary, international consortium; interest in cultural heritage applications is welcome.
• Excellent command of written and spoken English; French is not required., Research Field Computer science

Education Level PhD or equivalent

Languages ENGLISH

Level Excellent

Research Field Computer science

Years of Research Experience None

Salary According to the CY Cergy Paris Université salary scale, commensurate with experience

ECHOES (European Cloud for Heritage OpEn Science, www.echoes-eccch.eu) is a large European project, funded by the European Commission and UKRI, that is building the European Collaborative Cloud for Cultural Heritage (ECCCH): a shared, distributed and federated digital platform giving heritage professionals and researchers access to data, scientific resources, and advanced digital tools. At the heart of the ECCCH lies a conceptual Knowledge Graph (KG), grounded in established semantic standards such as CIDOC-CRM, that supports the description, annotation, integration and search of cultural heritage assets and underpins the Heritage Digital Twin (HDT) and the emerging Digital Commons Knowledge Base. By design, the ECCCH is not a single centralised repository: data, metadata and computing resources remain distributed across cultural heritage institutions, national nodes and connected European infrastructures (e.g., the Data Space for Cultural Heritage, Europeana, EOSC). This raises fundamental research questions on how to learn from, reason over, and enrich knowledge graphs that are physically and administratively distributed, without requiring the centralisation of the underlying data. Scientific objectives of the position The postdoctoral researcher will investigate distributed machine learning methods that operate over distributed and federated knowledge graphs, contributing both novel scientific results and concrete components for the ECHOES ecosystem. Research directions include (non-exhaustively): * Federated graph representation learning: extending knowledge graph embedding models (e.g., translational, bilinear and rotational families) and relational graph neural networks (e.g., R-GCN and successors) to settings where triples and subgraphs are partitioned across autonomous sources and cannot be centralised. * Learning under statistical and semantic heterogeneity: handling non-IID data distributions, heterogeneous schemas and vocabularies, partial overlaps and varying data quality across the participating cultural heritage institutions, including schema/ontology alignment as part of the learning loop. * Cross-graph inference tasks: distributed entity resolution and entity alignment, link prediction and knowledge graph completion across federated KGs, and their use for semantic enrichment of Heritage Digital Twins. * Communication-efficient and privacy-aware protocols: aggregation strategies for federated optimisation (e.g., FedAvg-style and personalised variants adapted to graph data), compression and sampling of graph structures, and privacy-preserving mechanisms (secure aggregation, differential privacy) respecting institutional data governance and intellectual property constraints. * Scalable distributed querying and analytics: interplay between federated SPARQL query processing, distributed graph partitioning, and learning pipelines over large, evolving KGs. * Evaluation on real cultural heritage data: benchmarking on datasets and use cases provided by ECHOES partners and sister projects, and integration of prototypes into the ECCCH technical backbone, released as open source. The successful candidate is expected to publish in leading venues in machine learning, data management and the Semantic Web (e.g., NeurIPS, ICML, ICLR, KDD, WWW, ISWC, ESWC, VLDB, EDBT), to contribute to project deliverables, and to participate in the supervision of PhD and MSc students working on related topics.